Pharmacy benefit manager turns to analytics to fight healthcare fraud

Fraud becoming more complex

In the industry, pharmacy benefit managers typically have built their analytics from the ground up internally, said Jo-Ellen Abou Nader, Prime's assistant vice president of fraud, waste and abuse operations. "The problem is fraud has gotten much more complex, and we need a big engine, as well as a partner in the industry to move this forward," Abou Nader said. "That's why we selected SAS. They have experience in the healthcare industry with IT, and their Fraud Framework was a great engine and opportunity for us to partner with them on building this out."

Stu Bradley

The SAS Fraud Framework, a cloud-based product hosted by SAS, enables Prime to manage the entire fraud protection lifecycle, said Stu Bradley, vice president of fraud and security intelligence at SAS. The product "has the capabilities to ingest, enrich, transform data. So we've got a big data management component which is inclusive of data quality. It gives us the ability to build and deploy rules all the way through to analytic models, leveraging things like machine learning and artificial intelligence, and deploy those models into a runtime environment such that we can execute it against claims from a healthcare perspective."

Before SAS entered the picture, Prime used rule-based analytics to scour its pharmacy claims for potential fraud. "The rules are very foundational to identify fraud that we knew of," Abou Nader said. "But where we're heading with SAS is being able to look across not only one client's claims, but we're able to pull all of the claims in for all our clients under Prime, all of our Blue Plans, to be able to look across the entire network of pharmacies and of members and physicians so that we're not working in a silo."

Jo-Ellen Abou Nader

For Prime, the SAS Fraud Framework will include more than 1,000 scenarios, or models, designed to identify the risk of healthcare fraud. The framework will be built out by SAS "so that all these scenarios trigger risk scores," Abou Nader said. "There are risk scores for each member, for each physician, for each pharmacy. That's how our analytics team will prioritize."

By integrating pharmacy and medical claims, Abou Nader said, "a scenario we may look at is pharmacy claims with no associated medical office visits-- [for example,] a pain management patient where they are on controlled substances but we see no office visits within the last six months. We're looking also at duplicate therapy across the pharmacy benefit and medical benefit.

"It's not that you have to look in each scenario," Abou Nader said. "SAS has done the job for us so it's not a needle in a haystack. They have identified where each physician, for example, triggers on each scenario."

Taking a holistic approach

What Bradley finds intriguing from an analytics perspective is the ability to look at data patterns across various insurance plans.

I like to say [fraud] is like whack-a-mole. As soon as we've caught the last fraud, here comes the next scenario.

By managing healthcare fraud detection at an individual level, the Blue Plans "have insight only into their own data and their own claims," Bradley said. "But if you can look holistically across those 22 plans, you have a much better opportunity to identify some of the more complex fraud schemes that may extend across multiple plans and be able to link that data together such that you can find and identify where you might have great risk. We call that consortium modeling."

This approach, he said, provides "better accuracy, better prioritization of risk and it reduces the false positives because you can look at, say, a specific pharmacy that is paying prescriptions across multiple plans and be able to look at their behavior holistically versus in their respective silos."

Because the Fraud Framework is hosted by SAS, implementation on Prime's part centered on ensuring data quality. "There has been a lot of discussion around pushing this amount of data to SAS," Abou Nader said. The focus particularly has been on "how they digest the information, because medical claims look very different than pharmacy claims. With this integration we've really had to focus on how the analytics will look and the outcomes we want, so it's really been a process to go through with SAS to get to where we need to be."

Don't overlook the importance of data

One example of the tandem approach Abou Nader mentioned centered on preserving the quality of the claims data under examination.

When you consider the overall lifecycle of fraud detection, Bradley said, "it's got to start with the data. We've got to be able to provide the appropriate analytic capabilities to build and deploy those models. You need to be able to execute and score every single transaction, and in this case it's claims."

Ensuring the quality of Prime's data was a joint effort, Bradley said. The effort starts "with the identification of the appropriate data sources. We leverage their domain experts who know a lot about Prime's data, in concert with our domain experts who know a lot about broader pharmacy claims data across multiple organizations and how we need to best structure that for fraud. Part of that is cross-pollinating those skill sets and defining how we're going to transform that data and get it into an analytical-ready format such that we can deliver [more quickly] analytics that is going to be more accurate, and serve up the data to the end users that is going to be in as valuable a format as possible." 

Fraud continually evolves

SAS has "gathered all of our data," Abou Nader said. "Now we're in the process of pushing all that data and defining all the analytics and the model scoring as it relates to our set of data. So it's quite a bit of lift on the SAS side."

She said Prime and SAS are targeting an October go-live date for the healthcare fraud framework. "Our goal is to be able to get cases out to test them. We're expecting to see results as early as September, when we will start getting alerts to test -- the actual alerts to say look at this member, look at this physician, look at this pharmacy, they are the highest risk. We will start getting some of those in the September, October timeframe to be able to test and go validate to say is this fraud or is this not fraud."

But this won't see the end of the work. Fraud is an evolution, Abou Nader said, and so is healthcare fraud data analytics modeling. "I like to say [fraud] is like whack-a-mole. As soon as we've caught the last fraud, here comes the next scenario.

"The great thing about SAS is this is a long-term partnership where we will continue to update the models with the latest fraud schemes that are happening."